ConvNetSharp Versions Save

• Update to .NET Core 3.1 / .NET Standard 2.1
• Cuda 10.0 / cuDNN cuDNN v7.6.4

• Remove unused regularization parameters (#137)
• Managed Cuda 9 1 (#136)
• Fix Dropout exception (#133)

Bug fixes:

• Wrong gradient computation in Flow when batch size > 1 #128

New features:

• Matrix multiplication / Transpose / RNN demo #129

Bug fixes:

• GPU memory improvements #121
• Make Shape simplier => always 4 dimensions #120
• Dropout is wrong in Flow #118

• Fix missing kernels

• Crash in backward pass of Dropout in GPU (#90)
• DropoutLayer Missing Method Exception (#92)
• Add Dropout in Core and Flow #96
• Assigment op (#97 )
• Sqrt op (#98 )
• Power op (#99)
• Adam optimizer (#100 )

Bug fix: o Missing kernels in resources (#85 )

• Enforce use of BuilderInstance rather than 'new Volume'
• Fix missing initialisation in regression layer
• netstandard 2.0

• DropOut layer #42
• Leaky ReLU #12

• Save and Load Serialization #67
• Some fixes for dropout layers #71
• Fix regression layer loss computation

• Fix several issues in ConvNetSharp.Flow
• Some perf improvements